95% of enterprise AI pilots fail. It's not the AI.

95% of enterprise AI pilots deliver zero measurable ROI. That's MIT research, not a disgruntled consultant with an axe to grind.

The takes I keep seeing in response: the models aren't ready. Vendors overpromised. AI is just hype dressed up in a suit. Understandable reactions. Wrong diagnosis.

MIT's finding is more uncomfortable than that. Companies using identical models, the same technology, the same vendors, are getting wildly different results. The model isn't the variable. Something else is.

Here's what I think that something else is.

Nobody defines what success looks like before the pilot starts. "Let's run a pilot and see what happens" sounds reasonable. In practice it's a way of avoiding the harder conversation about what you actually need the AI to do, and how you'd know if it was doing it. Without that, you can't measure anything. Months pass. Nothing is obviously working. Someone pulls the budget. The pilot dies not because it failed but because nobody agreed on what passing looked like.

The process underneath is already broken. AI doesn't fix broken processes. It accelerates them. If your data is a mess, your workflows are undocumented, and nobody has clear ownership of the output, introducing AI doesn't solve any of that. It just makes the problems move faster. Most pilots don't budget for the cleanup work that needs to happen first. That work isn't glamorous, it doesn't get a launch announcement, but it's quietly most of the job.

Nobody manages the change. The team whose day-to-day workflow gets rewritten. The manager who finds out about the project secondhand. The senior stakeholder who approved the budget in a slide deck but never really understood what they were approving. None of those are technology problems. All of them will kill a deployment. Every time.

The 95% failure rate is not evidence that AI doesn't work. It's evidence that most organisations haven't worked out how to implement it properly. That's a different problem, and it has a different solution.

The organisations actually getting value from AI right now aren't doing anything exotic. They define what good looks like before they start. They fix their data and processes before a model goes anywhere near them. And they treat adoption as a people problem, not a technology one, from day one.

That last part is where most of it gets lost.

These are my personal views and do not reflect the position of my employer or any organisation I am affiliated with.